--- title: "Build or buy? Lean towards build" description: "Build or buy is an old question that companies and startups are facing again. More and more, my answer is build. Why, and when buying still makes sense." date: 2026-11-07 image: "/uploads/2026/11/build-or-buy.jpg" categories: - "AI" - "Method" - "Innovation" --- Build or buy is one of the oldest questions in software, and a lot of companies and startups are facing it again right now. For my own needs, I keep landing on build. Not because I enjoy reinventing things, but because building keeps getting easier while buying costs the same as it always did. ## Even Photoshop is fair game A good sign of where things are going is [Craft Apps](https://getartcraft.com/apps), a growing set of free, open source apps from the team behind ArtCraft. They're rebuilding the tools everyone has paid for, one by one: an image editor in the spirit of Photoshop, vector apps like Illustrator, a Figma-style design tool, a word processor, a spreadsheet, presentations, even CAD. Most are still in development and a few are in early alpha. But the direction is clear, and new apps keep getting added. Software that took decades and thousands of people to build is being recreated and given away. When that's possible, the price of most tools stops telling you much about how hard they were to make. ## My own analytics I wanted to see what happens across all my projects in one place. Page views, events, signups, and which AI agents visit my sites, which normal analytics miss because agents don't run JavaScript. The usual way would be to look for a tool that does most of that, accept the gaps, and pay for it every month. Instead I built Appscope. It tracks exactly what I care about, pulls in revenue, sends me a summary every Monday, and lets my coding agents read the numbers directly. It costs me nothing to run, and when I want something new I add it the same day. It's one of the projects in my [Lab](/lab). The freedom to reshape it whenever my needs change is the part that's hard to get from bought software. ## Why companies still buy I see both sides with the companies I work with. Teams that build quickly and teams that hesitate, asking whether they can really do this. The default is still to buy what someone else bought at a high price, almost as if software isn't real unless it's expensive. I understand why. At a big company it's not your money, there might already be a budget for it, and nobody gets blamed for buying the big product everyone else uses. If you build your own and something goes wrong, it's on you. That fear is reasonable, and it's also what keeps a lot of expensive contracts alive. ## A spreadsheet that got its own tool One example stuck with me. An internal marketing team had outgrown a complex Excel sheet and was choosing between Monday.com and building its own tool. Setting Monday.com up for their requirements and keeping it running would have been expensive, and they would have had to wait for the vendor to find time. They built their own instead. It was done quickly, and they can change it whenever they want. They're not alone. In [Retool's build vs buy report](https://retool.com/blog/ai-build-vs-buy-report-2026) from February, 35 percent of the teams asked had already replaced at least one SaaS tool with something they built, and 78 percent expected to build more this year. Project management tools were among the things being replaced. It's a survey of Retool's own users, so it leans towards builders, but bigger companies are saying the same thing out loud. Palantir said on its earnings call in May that it had replaced its old, expensive CRM with one it built itself in a few months. ## The final boss Then there's enterprise software, and SAP is the final boss of it. Licensing advisers put SAP's yearly support fee at around 22 percent of the licence value, which for a large company means millions every year just to keep the system alive and maintained. Today a company could build its own version, mapped exactly to how it works. It would need developers to maintain it, but not many. One or two who work fully with AI could probably do it. Even with very expensive developers, it would cost a fraction of the fees, and it could be shaped far more than any configured SAP install. I wouldn't want to work on that project myself, but the opportunity is real, and smaller versions of it are everywhere. ## Paying for a wrapper I also see a lot of companies buying expensive AI wrappers. A nice interface on top of an image model, a writing model or a search model that the vendor doesn't own either. Going straight to the model's API and building your own interface on top is completely feasible today, at a fraction of the price, and you get exactly the workflow your team needs. ## One machine instead of many vendors The biggest advantage of building comes later. Tools you build yourself can be connected, share the same data and automate the work between them, until they start to run larger parts of the business as [one machine](/thoughts/build-the-machine-not-the-thing). That's much harder with software from different vendors, even when they have integrations. You're rarely free to change things the way you want, and asking the vendor to change them is expensive and slow. ## When to buy anyway If building costs almost nothing and keeping it running is cheap, the useful question is what the vendor has that you can't get by writing code. These are the questions I'd go through before building something myself. - **Does it get better because others use it?** A social network, a marketplace or a booking platform is valuable because of who's already on it. Your own version starts empty. - **Is the value in data the vendor has collected?** Maps, credit checks, market data or weather. The software is the easy part, and years of data can't be made in an afternoon. - **Does it depend on trust built over years?** Sending email is simple, but landing in the inbox depends on a reputation with the big email providers that takes years to earn. - **Does it need certifications, licences or legal standing?** Payments, BankID, e-signatures that hold up legally, or anything that has to pass audits. A vendor has already been through all of that. - **Does it involve physical steps?** Robots, machines, sensors or anything that has to move things in the real world. Hardware is still slow and expensive to make, so buy it, at least for now, and build the software that runs it. - **Does it move faster than you could follow?** Frontier AI models are the clearest case. Training one costs a fortune and the big labs release something new every few months. Smaller, specialised models are starting to make sense for some companies, but the frontier belongs to the few who can afford it. - **Is it already cheap and good enough?** If a tool costs little, does what you need and you'd never want to change it, there's no reason to rebuild it just because you can. If every answer is no, and you're paying a lot for something you'd like to shape your own way, I'd take the chance and try building it first.